Poincaré–Lindstedt Method

Almost every oscillator in physics is secretly nonlinear. A pendulum swung wide, a mass on a stiffening spring, a molecule stretched past its harmonic well, a planet tugged by its neighbours — each is a harmonic oscillator with a small correction of size $\epsilon\ll1$ . The instinct is to expand the motion as $x=x_0+\epsilon x_1+\cdots$ and match powers of $\epsilon$ . This regular perturbation series is correct term by term, yet it fails catastrophically: every order past the first is a harmonic oscillator driven at its own natural frequency, and that resonance breeds secular terms proportional to $t\sin t$ that grow without bound. The true motion is a bounded, periodic wiggle; the approximation to it runs off to infinity. Formal correctness at fixed $t$ is worthless once $\epsilon t=O(1)$ — after a few periods the series describes nothing real. The Poincaré–Lindstedt method is the cure, and its insight is physical, not algebraic. A nonlinear restoring force makes the period depend on amplitude: swing the pendulum wider and it runs slower. No expansion that hard-codes the linear frequency can represent an amplitude-dependent one — so it fakes the shift with the only tool a Taylor series has, the ruinous secular term. Poincaré–Lindstedt gives the expansion an honest knob: it strains the clock, $\tau=\omega t$ , and expands the unknown frequency $\omega=\omega_0+\epsilon\omega_1+\cdots$ alongside the displacement, spending the freedom in each $\omega_k$ to annihilate the resonant forcing order by order. The method is the simplest of the singular-perturbation tools, the historical seed of averaging and multiple scales, and the cleanest demonstration that in asymptotics the goal is uniform validity, not mere term-by-term truth.

Key takeaways

Weakly nonlinear oscillators $\ddot x+x+\epsilon f(x,\dot x)=0$ resist naive perturbation theory: higher orders are resonantly driven and produce secular terms $t\sin t$ that grow without bound.
The fix is to strain time, $\tau=\omega t$ , and expand the frequency $\omega=\omega_0+\epsilon\omega_1+\cdots$ together with $x=x_0+\epsilon x_1+\cdots$ .
Each frequency correction $\omega_k$ is fixed by a solvability (Fredholm) condition: the forcing must be orthogonal to the resonant modes $\cos\tau,\sin\tau$ .
For the Duffing oscillator $\ddot x+x+\epsilon x^3=0$ of amplitude $a$ this gives the amplitude–frequency relation $\omega=1+\tfrac38\epsilon a^2+O(\epsilon^2)$ — a hardening spring speeds up as it swings wider.
The same machinery yields the pendulum's circular-error period $T=T_0(1+\theta_0^2/16+\cdots)$ and fixes the van der Pol limit-cycle amplitude to $a=2$ .
It applies to periodic orbits of conservative or self-excited systems; for modulated amplitudes one graduates to multiple scales or averaging.
Key idea: Strain the time variable — write $\tau=\omega t$ and expand $\omega=\omega_0+\epsilon\omega_1+\cdots$ alongside $x=x_0+\epsilon x_1+\cdots$ . Spend the freedom in each $\omega_k$ to kill the resonant (secular) forcing, guaranteeing a bounded periodic solution.

The core insight

Strain the time variable: write $\tau=\omega t$ and expand $\omega=\omega_0+\epsilon\omega_1+\cdots$ alongside $x=x_0+\epsilon x_1+\cdots$ . The freedom in each $\omega_k$ is spent eliminating resonant (secular) forcing, guaranteeing a bounded periodic solution.

Where this lives today

The amplitude–frequency backbone curve is measured daily in the laboratory. Micro- and nano-electromechanical resonators (MEMS/NEMS) — the timing elements in phones, the mass sensors that weigh single molecules — are driven hard enough that their restoring force stiffens, tracing out exactly the Duffing backbone this method predicts; engineers read the tilt of the resonance peak to extract the nonlinear coefficient and stabilise frequency references. In particle accelerators the same mathematics is the amplitude-dependent tune shift: a particle with larger betatron amplitude sees a slightly different focusing frequency, and controlling that shift — a Poincaré–Lindstedt calculation in disguise — keeps beams stable over billions of turns. Nonlinear optics inherits the third-harmonic term directly: a Kerr medium generates light at $3\omega$ with amplitude scaling as the input cubed, the optical echo of equation (6). And at the deepest level the method's small-divisor troubles matured into the Kolmogorov–Arnold–Moser theorem — Poincaré's question about solar-system stability answered: most strained-frequency series survive, a measure-zero resonant set does not, and that razor's edge between order and chaos is where much of dynamical-systems research still lives.

The physical picture

Watch what the naive series actually does. Its leading term is the honest harmonic motion $a\cos t$ , bounded and periodic. But the next term contains $t\sin t$ : an oscillation whose amplitude ramps up linearly with time. Order by order the powers of $t$ pile higher, and the sum — meant to approximate a bounded swing — instead spirals outward inside an ever-widening envelope. The true solution and the approximation share the same first wiggle and then part ways; within a handful of periods the naive curve describes a runaway that does not exist. The first figure freezes this divorce: the bounded Poincaré–Lindstedt solution against the naive series climbing its secular envelope.

Figure

Naive expansion versus Poincaré–Lindstedt for the Duffing oscillator ẍ+x+εx³=0 with ε=0.14 and a=1: the blue Poincaré–Lindstedt solution a cos ωt stays bounded within ±1, while the red naive series climbs its secular envelope ±√(1+(3εt/8)²) and diverges — the two agree only for εt≪1 near the start

Why must the frequency be part of the story? Expand the shifted cosine $\cos((1+\delta)t)=\cos t-\delta\,t\sin t+\tfrac12\delta^2 t^2\cos t-\cdots$ and the entire ladder of secular terms falls out of a perfectly bounded function. The naive series is not wrong about the physics — it is re-expanding a frequency shift into powers of $t$ . Poincaré–Lindstedt refuses that Taylor expansion, keeping $\omega$ exact inside the argument: resummation by foresight, in which the most dangerous subseries is pre-summed into a slow modulation of the phase. The payoff is measurable — the amplitude–frequency backbone curve the second figure plots for the Duffing oscillator.

Figure

Backbone curve of the Duffing oscillator: the amplitude–frequency relation ω=1+(3/8)εa² at ε=0.1 — the small-amplitude frequency ω=1 rises to ω=1.0375 at a=1 and ω=1.15 at a=2, so a hardening spring speeds up as its swing grows

The core equations

The archetype of the whole subject is the Duffing oscillator — a unit-frequency harmonic oscillator with a small cubic restoring force. It is conservative, so its energy and hence its amplitude $a$ are fixed by the initial data; that makes $a$ a free parameter the frequency will end up depending on.

\ddot{x}+x+\epsilon\,x^3=0,\qquad x(0)=a,\ \dot{x}(0)=0 \qquad (1)

where: $x(t)$ — displacement; $\epsilon$ — small nonlinearity parameter, $\epsilon\ll1$ ; $a$ — initial amplitude, a free constant for this conservative system; $\dot{x}$ — $\mathrm{d}x/\mathrm{d}t$ . Reading: a linear spring ( $\omega=1$ ) stiffened by a cubic term whose strength is $\epsilon$ ; a hardening spring for $\epsilon>0$ , softening for $\epsilon<0$ . Feed this to the naive regular expansion and out comes the secular disaster. The correct first-order result — verify the coefficients yourself — is the bounded harmonic plus a term that grows linearly in $t$ .

x(t)=a\cos t-\frac{\epsilon a^3}{8}\bigl(\,3\,t\sin t+\tfrac14\cos t-\tfrac14\cos 3t\,\bigr)+O(\epsilon^2) \qquad (2)

where: $3\,t\sin t$ — the secular term; its amplitude grows without bound, ruining the approximation once $\epsilon t\gtrsim1$ ; $\cos 3t$ — an honest, bounded third harmonic. Reading: the only pathological piece is $t\sin t$ ; everything else is legitimate. The repair is to stop treating time as fixed. Introduce a strained clock $\tau=\omega t$ and let the true frequency $\omega$ be unknown, expanded in $\epsilon$ right beside the displacement.

\tau=\omega t,\qquad \omega=\omega_0+\epsilon\omega_1+\epsilon^2\omega_2+\cdots,\quad \omega_0=1 \qquad (3)

where: $\tau$ — strained (fast) time; $\omega$ — the true nonlinear frequency, unknown; $\omega_k$ — frequency corrections, each fixed by killing a secular term. Reading: $\tau$ advances by exactly $2\pi$ over one true period, so the solution is $2\pi$ -periodic in $\tau$ by construction. Under $\tau=\omega t$ each time derivative brings down a factor $\omega$ , so $\mathrm{d}^2/\mathrm{d}t^2=\omega^2\,\mathrm{d}^2/\mathrm{d}\tau^2$ . The Duffing equation becomes an equation in $\tau$ with $\omega^2$ sitting in front of the acceleration.

\omega^2\,x''+x+\epsilon x^3=0,\qquad (\,'\,)\equiv\frac{\mathrm{d}}{\mathrm{d}\tau},\qquad \omega^2=1+2\epsilon\omega_1+\epsilon^2(\omega_1^2+2\omega_2)+\cdots \qquad (4)

where: $x''$ — second derivative with respect to $\tau$ ; $\omega^2$ — expanded so that $2\omega_1$ appears at order $\epsilon$ . Reading: the extra $2\epsilon\omega_1 x_0''$ is exactly the counter-term that cancels the resonant part of the cubic forcing. Demanding no secular term at first order fixes $\omega_1=\tfrac38 a^2$ , giving the central result of the method — the amplitude–frequency relation, or backbone curve.

\omega=1+\tfrac{3}{8}\,\epsilon a^2+O(\epsilon^2) \qquad (5)

where: $\tfrac38\epsilon a^2$ — the leading frequency shift. Reading: the frequency rises with amplitude for a hardening spring ( $\epsilon>0$ ) and falls for a softening one — the defining signature of a nonlinear oscillator, and something no fixed-frequency expansion can express. With $\omega_1$ chosen, the remaining first-order equation is non-resonant and solves cleanly, leaving a bounded periodic answer with a third harmonic — the fingerprint of a cubic nonlinearity.

x(t)=a\cos\bigl(\omega t\bigr)+\frac{\epsilon a^3}{32}\bigl(\cos 3\omega t-\cos\omega t\bigr)+O(\epsilon^2) \qquad (6)

where: $\cos 3\omega t$ — third-harmonic generation from the cubic term; $\omega$ — the corrected frequency $1+\tfrac38\epsilon a^2$ . Reading: bounded for all $t$ — uniformly valid — because the frequency shift now lives inside the cosine argument, not out front as a growing coefficient. The bookkeeping device that fixes each $\omega_k$ is a solvability (Fredholm) condition. Each order is a driven oscillator $x_n''+x_n=F_n(\tau)$ ; a bounded $2\pi$ -periodic solution exists only if the forcing carries no component along the resonant modes $\cos\tau,\sin\tau$ .

\int_0^{2\pi}F_n(\tau)\cos\tau\,\mathrm{d}\tau=0,\qquad \int_0^{2\pi}F_n(\tau)\sin\tau\,\mathrm{d}\tau=0 \qquad (7)

where: $F_n$ — the forcing assembled at order $n$ ; $\cos\tau,\sin\tau$ — the homogeneous solutions (the kernel of $\mathrm{d}^2/\mathrm{d}\tau^2+1$ ). Reading: since $F_n$ depends linearly on $\omega_n$ , each integral is one linear equation that pins $\omega_n$ down and guarantees boundedness. The same idea, applied to the pendulum, recovers a classical result. Writing $\sin\theta\approx\theta-\theta^3/6$ turns the pendulum into a softening Duffing oscillator whose amplitude $\theta_0$ plays the role of the small parameter.

\ddot{\theta}+\frac{g}{\ell}\sin\theta=0\ \Longrightarrow\ T=\frac{2\pi}{\omega_0}\Bigl(1+\tfrac{1}{16}\theta_0^2+O(\theta_0^4)\Bigr) \qquad (8)

where: $\theta_0$ — angular amplitude; $\omega_0=\sqrt{g/\ell}$ — small-oscillation frequency. Reading: a wider swing means a longer period — the pendulum's circular error — captured exactly by a frequency shift $\omega=\omega_0(1-\theta_0^2/16)$ .

Derivation AWhere the secular terms come from

The purpose here is to watch the naive expansion fail in slow motion, so the cure looks inevitable rather than magical. We substitute the plain series $x=x_0+\epsilon x_1+\cdots$ into the Duffing equation with the clock untouched and trace exactly which piece of the cubic forcing plants the runaway $t\sin t$ . The failure is structural, not a slip: resonance is built into any expansion that holds the frequency fixed while the physics wants it to move.

Step 1. Substitute $x=x_0+\epsilon x_1+\cdots$ into $\ddot x+x+\epsilon x^3=0$ and collect powers of $\epsilon$ . At order $\epsilon^0$ the equation is $\ddot x_0+x_0=0$ with $x_0(0)=a,\ \dot x_0(0)=0$ , so $x_0=a\cos t$ . Why: at leading order the cubic term is invisible and the motion is purely harmonic — the honest starting point.

Step 2. At order $\epsilon^1$ the cubic term evaluated on $x_0$ becomes the forcing: $\ddot x_1+x_1=-x_0^3=-a^3\cos^3 t$ . Why: the nonlinearity, small but nonzero, feeds the leading motion back in as a drive on the next correction — this is where all the trouble enters.

Step 3. Reduce the cube with the identity $\cos^3 t=\tfrac34\cos t+\tfrac14\cos 3t$ , so $\ddot x_1+x_1=-\tfrac34 a^3\cos t-\tfrac14 a^3\cos 3t$ . Why: the drive splits into a piece at the natural frequency ( $\cos t$ ) and a piece at three times it — only the first is dangerous.

Step 4. The $\cos t$ drive is exactly on resonance. A forced oscillator $\ddot x_1+x_1=A\cos t$ has particular solution $\tfrac{A}{2}t\sin t$ , so the $-\tfrac34 a^3\cos t$ piece produces $-\tfrac38 a^3\,t\sin t$ — an amplitude that grows linearly forever. Why: driving an undamped oscillator at its own frequency pumps energy in without limit; the $t\sin t$ is that resonant pumping, and it is the secular catastrophe.

Step 5. Solve the non-resonant $\cos 3t$ drive and fix the homogeneous constants from $x_1(0)=0,\ \dot x_1(0)=0$ , giving $x_1=-\tfrac38 a^3 t\sin t-\tfrac{a^3}{32}\cos t+\tfrac{a^3}{32}\cos 3t$ . Why: this is equation (2) in full; the bounded harmonics are fine, but the leading $t\sin t$ dooms the series for $\epsilon t\gtrsim1$ no matter how many orders we keep. *Verified: $\cos^3 t=\tfrac34\cos t+\tfrac14\cos 3t$ ; resonant particular solution $\tfrac{A}{2}t\sin t$ with $A=-\tfrac34a^3$ gives the $-\tfrac38a^3\,t\sin t$ secular term (symbolic check).*

Go deeper · Derivation B · 2 minThe Duffing oscillator solved without secular terms

Derivation BThe Duffing oscillator solved without secular terms

Now we run the same problem with the strained clock and reap a uniformly valid answer. Substitute both expansions — $x=x_0+\epsilon x_1+\cdots$ and $\omega^2=1+2\epsilon\omega_1+\cdots$ — into the strained equation (4), collect orders, and at first order use the free constant $\omega_1$ to annihilate the resonant $\cos\tau$ forcing before it breeds a secular term. The frequency correction is no afterthought; it is the specific counter-term the solvability condition demands, and choosing it is the whole method in one move.

Step 1. Insert $x=x_0+\epsilon x_1+\cdots$ and $\omega^2=1+2\epsilon\omega_1+\cdots$ into $\omega^2 x''+x+\epsilon x^3=0$ . At order $\epsilon^0$ : $x_0''+x_0=0$ with $x_0(0)=a,\ x_0'(0)=0$ , so $x_0=a\cos\tau$ . Why: in the strained clock the leading motion is still simple harmonic — but now it is periodic in $\tau$ , not $t$ .

Step 2. At order $\epsilon^1$ two contributions appear — the $2\omega_1$ from $\omega^2$ acting on $x_0''$ , and the cubic term acting on $x_0$ : $x_1''+x_1=-2\omega_1 x_0''-x_0^3=2\omega_1 a\cos\tau-a^3\cos^3\tau$ . Why: straining the clock has injected a brand-new $\cos\tau$ term proportional to $\omega_1$ — the adjustable knob that was missing from the naive route.

Step 3. Expand the cube, $-a^3\cos^3\tau=-\tfrac34 a^3\cos\tau-\tfrac14 a^3\cos 3\tau$ , so the total coefficient of the resonant $\cos\tau$ is $2\omega_1 a-\tfrac34 a^3$ . Why: this coefficient is precisely the resonant Fourier component the solvability condition (7) will test.

Step 4. Demand no secular term: set the resonant coefficient to zero, $2\omega_1 a-\tfrac34 a^3=0$ , giving $\omega_1=\tfrac38 a^2$ . Why: this is the solvability condition in action — one algebraic equation that both removes the runaway and delivers the amplitude–frequency relation $\omega=1+\tfrac38\epsilon a^2$ .

Step 5. With resonance gone, the leftover equation $x_1''+x_1=-\tfrac14 a^3\cos 3\tau$ is non-resonant; its particular solution is $\tfrac{a^3}{32}\cos 3\tau$ , and fixing the homogeneous constant by $x_1(0)=0$ gives $x_1=\tfrac{a^3}{32}(\cos 3\tau-\cos\tau)$ . Why: a $\cos 3\tau$ drive on a unit oscillator responds with amplitude $1/(1-9)=-\tfrac18$ , bounded and safe — the third harmonic is generated but never grows.

Step 6. Restore $\tau=\omega t$ and assemble: $x=a\cos\omega t+\tfrac{\epsilon a^3}{32}(\cos 3\omega t-\cos\omega t)+O(\epsilon^2)$ with $\omega=1+\tfrac38\epsilon a^2$ . Why: bounded and $2\pi/\omega$ -periodic, this is uniformly valid for all $t$ to this order — the frequency shift now lives safely inside the cosine. *Verified: matches Bender & Orszag Ch. 11 and Stone & Goldbart; $\omega_1=\tfrac38 a^2$ , third-harmonic coefficient $a^3/32$ , and $x_1(0)=0$ reproduced symbolically.*

Go deeper · Derivation C · 1 minThe van der Pol limit cycle from a solvability condition

Derivation CThe van der Pol limit cycle from a solvability condition

The van der Pol oscillator turns the method inside out. It is self-excited — small oscillations grow, large ones are damped — so it settles onto a unique attracting limit cycle whose amplitude the equation itself selects. Here the solvability condition does double duty: one component fixes the frequency, the other the amplitude. We strain time as before and read both out of the first-order resonance, discovering that periodicity alone forces $a=2$ .

Step 1. Take $\ddot x+x=\epsilon(1-x^2)\dot x$ , strain $\tau=\omega t$ with $\omega=1+\epsilon\omega_1+\cdots$ , and at order $\epsilon^0$ get $x_0''+x_0=0$ ; choose the phase so $x_0=a\cos\tau$ with no $\sin\tau$ . Why: fixing the phase removes a redundant constant and makes the resonance conditions clean.

Step 2. At order $\epsilon^1$ : $x_1''+x_1=-2\omega_1 x_0''+(1-x_0^2)x_0'$ , with $x_0'=-a\sin\tau$ . Why: both the frequency counter-term and the nonlinear damping now drive the first correction.

Step 3. Reduce the damping term: $(1-a^2\cos^2\tau)(-a\sin\tau)=-a\bigl(1-\tfrac{a^2}{2}\bigr)\sin\tau+\tfrac{a^3}{4}\sin 3\tau$ . Combined with $-2\omega_1 x_0''=2\omega_1 a\cos\tau$ , the resonant coefficients are $2\omega_1 a$ on $\cos\tau$ and $-a\bigl(1-\tfrac{a^2}{4}\bigr)$ on $\sin\tau$ . Why: a self-excited drive is odd in velocity, so it lands on $\sin\tau$ — a channel the conservative Duffing problem never touched.

Step 4. Impose both solvability conditions. From $\cos\tau$ : $2\omega_1 a=0\Rightarrow\omega_1=0$ . From $\sin\tau$ : $1-\tfrac{a^2}{4}=0\Rightarrow a=2$ . Why: the $\sin\tau$ condition is not about frequency at all — it selects the amplitude, revealing the limit cycle; the frequency correction is pushed to $O(\epsilon^2)$ , where it evaluates to $\omega=1-\tfrac1{16}\epsilon^2+\cdots$ .

Step 5. Conclude that the van der Pol oscillator relaxes to a periodic orbit of radius $a=2$ at leading order, independent of the starting amplitude. Why: the same Fredholm machinery that removed secular growth in Duffing here doubles as the amplitude-selection rule of a self-sustained oscillation. *Verified: standard van der Pol limit-cycle amplitude $a\to2$ and $\omega=1-\tfrac1{16}\epsilon^2+\cdots$ (Nayfeh, Perturbation Methods); $\sin\tau$ coefficient vanishes at $a=2$ (symbolic check).*

Limiting cases

The behaviour of the method depends sharply on the structure of the nonlinearity and on whether energy is conserved.

Case	Result	Meaning
Conservative, $\epsilon\to0$	$\omega=\omega_0+\epsilon\omega_1+\cdots$ , $a$ free	For Hamiltonian oscillators (Duffing, pendulum) the amplitude is a free constant of integration; only the frequency is corrected, and the orbit is genuinely two-sided periodic.
Self-excited (van der Pol)	$a=2$ , $\omega=1+O(\epsilon^2)$	The amplitude is not free: the $\sin\tau$ solvability condition fixes the limit-cycle radius at leading order while the frequency shift is pushed to second order.
Linear limit $\epsilon\to0$	$\omega\to\omega_0$ , $x\to a\cos\omega_0 t$	The correction switches off smoothly and the exact harmonic oscillator is recovered — the method reduces to ordinary linear theory.
Strong nonlinearity $\epsilon=O(1)$	series unreliable	The expansion need not converge or even remain usefully asymptotic; one turns to exact elliptic-function solutions or direct numerics.
Pendulum, moderate amplitude	$T=\tfrac{2\pi}{\omega_0}\bigl(1+\tfrac1{16}\theta_0^2+\cdots\bigr)$	Expanding $\sin\theta\approx\theta-\theta^3/6$ makes the pendulum a softening Duffing oscillator and recovers the classic circular-error correction.

Common trapExpanding the displacement but forgetting to expand the frequency

The single most common mistake is to strain nothing and expand only $x$ . That regular series is formally valid — correct to every order in $\epsilon$ at fixed $t$ — and useless, because its terms grow like $t^n$ and diverge from the true bounded motion once $\epsilon t\sim1$ . Term-by-term correctness is a trap: asymptotic truth at fixed $t$ is necessary but nowhere near sufficient, and physics lives at $\epsilon t=O(1)$ , over many periods, exactly where the naive series dies. Two further cautions. First, the method presumes a periodic orbit; for a self-excited system the amplitude is not free but fixed by a solvability condition. Second, the frequency shift is $O(\epsilon)$ only for a conservative nonlinearity like $x^3$ ; for van der Pol it is $O(\epsilon^2)$ , and assuming otherwise gives a spurious first-order frequency.

Worked examples

1 · The Duffing backbone — frequency, period, and harmonic content (fully worked)

A hardening Duffing oscillator $\ddot x+x+\epsilon x^3=0$ has $\epsilon=0.1$ . Find its frequency, period, and third-harmonic amplitude for amplitudes $a=1$ and $a=2$ , and compare to the linear oscillator.

Step 1. Frequency from the backbone relation (5): for $a=1$ , $\omega=1+\tfrac38(0.1)(1)^2=1+0.0375=\mathbf{1.0375}$ . For $a=2$ , $\omega=1+\tfrac38(0.1)(2)^2=1+0.15=\mathbf{1.15}$ . The larger swing runs measurably faster — the hallmark of a hardening spring.

Step 2. Period $T=2\pi/\omega$ . For $a=1$ : $T=6.2832/1.0375=\mathbf{6.056}$ , versus $2\pi=6.283$ for the linear oscillator — about $3.6\%$ shorter. For $a=2$ : $T=6.2832/1.15=\mathbf{5.464}$ , some $13\%$ shorter. The period is not a fixed clock; it tightens as the amplitude grows.

Step 3. Third-harmonic amplitude from (6) is $\epsilon a^3/32$ . For $a=1$ : $0.1\times1/32=\mathbf{0.0031}$ , about $0.3\%$ of the fundamental. For $a=2$ : $0.1\times8/32=\mathbf{0.025}$ , about $1.25\%$ of the fundamental $a=2$ . The waveform is nearly sinusoidal but carries a small $3\omega$ overtone that grows as $a^3$ — the classical signature of a cubic nonlinearity, and the origin of third-harmonic generation in nonlinear optics and electronics.

Step 4. Sanity-check the ordering: the fractional frequency and waveform corrections are both $O(\epsilon a^2)$ , small only while $\epsilon a^2\ll1$ . At $a=2,\ \epsilon=0.1$ we have $\epsilon a^2=0.4$ — not tiny, which is why the $15\%$ shift at $a=2$ is where a first-order truncation starts to strain and $\omega_2$ would matter. Frequency and period harden with amplitude; third-harmonic content grows as a³.

2 · The pendulum's circular error (fully worked)

A pendulum is released from rest at $\theta_0=30^\circ$ . By how much does its period exceed the small-angle value, and how does that compare with the exact answer?

Step 1. Convert the amplitude: $\theta_0=30^\circ=0.5236$ rad. The Poincaré–Lindstedt correction (8) gives the fractional period increase $T/T_0=1+\theta_0^2/16$ .

Step 2. Evaluate: $\theta_0^2=0.2742$ , so $T/T_0=1+0.2742/16=1+0.01714=\mathbf{1.0171}$ . The period is about $\mathbf{1.71\%}$ longer than the small-angle $T_0=2\pi\sqrt{\ell/g}$ — a real, measurable effect that made pendulum-clock accuracy an eighteenth-century obsession.

Step 3. Compare with the exact period, the complete elliptic integral $T/T_0=(2/\pi)K(\sin^2(\theta_0/2))=1.01741$ , so the Poincaré–Lindstedt value $1.01713$ is low by only $0.027\%$ — three-decimal agreement from a single frequency correction.

Step 4. Push to $\theta_0=60^\circ=1.047$ rad: the formula gives $T/T_0=1.0685$ against the exact $1.0732$ , now off by $0.44\%$ . The leading correction degrades gracefully as the swing widens, and the next term $+\tfrac{11}{3072}\theta_0^4$ recovers most of the gap. At 30° the swing is 1.71% slow, matching the exact result to three decimals.

Physics of the portrait

The method was born twice — once as a practical trick in the hands of an astronomer straining planetary series, and once as a rigorous theory in the hands of the mathematician who taught the world what an asymptotic series really is.

Origins · 1 minHenri Poincaré and how this was discovered

Henri Poincaré

1854–1912 · celestial mechanics and the qualitative theory of dynamical systems

Henri Poincaré was the last universalist of mathematics, and the method that bears his name is a small corner of a vast campaign: to understand the solar system's stability without solving it. In the 1880s the reigning tool of celestial mechanics was the trigonometric series of Delaunay and Lindstedt, plagued by secular and small-divisor terms that grew without bound. Anders Lindstedt had shown in 1882 that tuning the frequencies could formally suppress the secular terms — but no one knew whether the resulting series meant anything. Poincaré's answer, across the three volumes of Les Méthodes Nouvelles de la Mécanique Céleste (1892–1899), was revolutionary on two fronts. First, he proved these series are generally divergent yet asymptotic — worthless if summed to the end, superbly accurate if truncated early — inventing the modern definition of an asymptotic expansion. Second, when the series failed he changed the question, asking not for formulae but for the qualitative shape of trajectories: fixed points, periodic orbits, and the tangled homoclinic webs he was first to glimpse, the seed of chaos theory. His small-divisor worries would resurface, decades later, as the KAM theorem — the deep reason the naive series diverges and the strained one, sometimes, does not.

History

1788

— Lagrange meets secular terms in the perturbation series of the planetary problem

1882

— Anders Lindstedt strains the time variable to suppress secular growth in celestial series

1892

— Poincaré's Les Méthodes Nouvelles puts perturbation series on a rigorous asymptotic footing

1918

— Georg Duffing analyzes the hardening-spring oscillator that now bears his name

1926

— van der Pol's relaxation-oscillation study gives the method its self-excited testbed

Going further

Texts. Bender & Orszag, Advanced Mathematical Methods for Scientists and Engineers, Ch. 11 (the canonical treatment) · Nayfeh, Perturbation Methods, Ch. 3 · Stone & Goldbart, Mathematics for Physics, chapter on perturbation methods · Arfken, Weber & Harris, Mathematical Methods for Physicists, 7th ed. · Riley, Hobson & Bence, Mathematical Methods for Physics and Engineering · Hassani, Mathematical Physics, for the Fredholm/self-adjoint background.

Provenance. The model problem and Duffing result ( $\omega=1+\tfrac38\epsilon a^2$ , third-harmonic term $\cos 3\omega t$ ) follow Bender & Orszag Ch. 11 and Stone & Goldbart; the van der Pol limit-cycle amplitude $a=2$ and $\omega=1-\tfrac1{16}\epsilon^2$ follow Nayfeh; the pendulum circular-error coefficient $1/16$ is the standard textbook value. Every number here was re-derived symbolically: the identity $\cos^3\tau=\tfrac34\cos\tau+\tfrac14\cos3\tau$ ; the resonant particular solution $\tfrac{A}{2}t\sin t$ giving the corrected naive secular term $-\tfrac38\epsilon a^3\,t\sin t$ (note: the migrated source dropped the factor of $3$ in the factored form — corrected in equation 2); the order- $\epsilon$ solvability $\omega_1=\tfrac38 a^2$ ; the van der Pol $\sin\tau$ coefficient vanishing at $a=2$ ; and the pendulum period $T/T_0=1.0171$ at $\theta_0=30^\circ$ against the exact elliptic value $1.0174$ .

Prerequisites. Asymptotic Series and Poincaré Expansions · Order Symbols and Gauge Functions · Linear ODEs and resonance. Leads to. Method of Multiple Scales · WKB Approximation · Bohr–Sommerfeld Quantization.

Ask the tutor: Want to see the second-order correction $\omega_2=-\tfrac{21}{256}a^4$ worked out, or watch the same solvability trick reappear inside the method of multiple scales?

What exactly is a secular term, and why is it fatal?

A secular term grows with time — typically $t\sin t$ , $t\cos t$ , or higher powers $t^n$ . The name is from astronomy (Latin saeculum, an age), where such terms seemed to predict slow, cumulative drifts in planetary orbits over centuries. They are fatal not because they are wrong at any fixed instant but because they destroy the uniformity of the approximation: the true motion of a conservative oscillator is bounded, so any unbounded term must eventually dominate and render the series meaningless. The breakdown sets in when the growing term becomes comparable to the leading one — when $\epsilon t=O(1)$ , after only a few periods. A secular term is a signal that the expansion is being asked to represent something it structurally cannot: a frequency shift.

Why expand the frequency instead of just working harder on the amplitude?

Because the effect the series fails to capture is a frequency shift. A nonlinear oscillator's period depends on its amplitude, and a series built on the fixed frequency $\omega_0$ has no vocabulary for that — so it mimics a shifted cosine $\cos(\omega t)$ by Taylor-expanding it in the small shift, generating exactly the secular ladder $\cos t-\delta t\sin t+\cdots$ . No amount of extra amplitude work fixes this: the amplitude corrections are bounded harmonics, while the disease lives in the phase. Expanding $\omega$ gives the series a direct handle on the phase, placing the frequency shift where it belongs — inside the cosine's argument — rather than faking it with growing coefficients out front.

How is Poincaré–Lindstedt related to the method of multiple scales, and to damping?

They are close cousins that agree on periodic problems and part company beyond them. Poincaré–Lindstedt uses a single strained time $\tau=\omega t$ and produces strictly periodic solutions with one amplitude-dependent frequency — ideal for conservative oscillators and limit cycles. Multiple scales introduces two or more time variables (a fast $t$ and a slow $T=\epsilon t$ ) and lets the amplitude and phase evolve slowly. That extra freedom is exactly what a linearly damped oscillator needs, since its amplitude decays and it has no periodic orbit for Poincaré–Lindstedt to correct; only self-excited damping like van der Pol's, which settles onto a limit cycle, stays within reach. For genuinely periodic motion both methods give the same frequency and waveform; for anything with amplitude dynamics, multiple scales is the general tool and Poincaré–Lindstedt the special case.

How do I get the higher-order corrections like ω₂?

Continue the same bookkeeping to order $\epsilon^2$ . Collecting all $\epsilon^2$ terms in the strained equation gives $x_2''+x_2=F_2(\tau)$ , whose forcing $F_2$ carries a resonant $\cos\tau$ component depending linearly on $\omega_2$ . Setting that coefficient to zero — the solvability condition again — fixes $\omega_2$ uniquely, just as before. For Duffing this yields $\omega_2=-\tfrac{21}{256}a^4$ , so $\omega=1+\tfrac38\epsilon a^2-\tfrac{21}{256}\epsilon^2 a^4+\cdots$ . The labour grows fast because the trigonometric reductions multiply, which is why people usually stop at first or second order and switch to numerics for high accuracy.

Is the Poincaré–Lindstedt series actually convergent?

Usually not — and that was Poincaré's discovery. For a one-degree-of-freedom conservative oscillator like Duffing the strained series has a finite radius of convergence and behaves well, but for the multi-frequency systems of celestial mechanics that motivated the method, the series are typically divergent. They are nonetheless asymptotic: truncated after a few terms they approximate the true motion superbly, but summed to the end they diverge, undone by small-divisor resonances between frequencies. This is not a defect but a structural fact — the same small divisors are why the KAM theorem must exclude a resonant set of initial conditions for invariant tori to survive. So Poincaré–Lindstedt gives an excellent finite-order approximation whose convergence you should never assume.

Why does the method carry two names?

Because it was created in two stages. Anders Lindstedt, a Swedish astronomer, introduced the frequency-straining trick in 1882 as a practical device to remove secular terms from the series used to compute planetary orbits — but he could not say whether the resulting series meant anything. Poincaré, a decade later in Les Méthodes Nouvelles, supplied the rigorous theory: he proved the series are asymptotic (not generally convergent), explained when and why the construction works, and embedded it in his theory of dynamical systems. The joint name honours both the trick and its justification — the astronomer's ingenuity and the mathematician's rigour.